Why app engine new price model is totally wrong

[Tapir]

Following I list the prices of main cloud hosting companies.

Hi, googlers, I just want to ask you a simple question: what is the
reason you think you should charge pretty much with pretty less
resource provided?

==============================================================================

App Engine backend price:
------------------------------------------------------------------
B1 128MB 600MHz $0.08 per hour
B2 256MB 1.2GHz $0.16 per hour
B4 512MB 2.4GHz $0.32 per hour
B8 1024MB 4.8GHz $0.64 per hour

On-demand Frontend Instances 24 Instance Hours $0.08 / hour
Reserved Frontend Instances $0.05 / hour

==============================================================================

EC2
----------------------------------------------------------------
Micro Instance: 613 MB, up to 2 ECUs (for short periodic bursts),
$0.02 per hour
Small Instance: 1.7 GB, 1 ECU, 160 GB of local instance storage, 32-
bit platform, $0.085 per hour
Large Instance 7.5 GB, 4 ECUs, 850 GB of local instance storage, 64-
bit platform, $0.34 per hour
Extra Large Instance 15 GB, 8 ECUs, 1690 GB of local instance storage,
64-bit platform, $0.68 per hour
High-Memory Extra Large Instance, 17.1 GB, 6.5 ECUs, 420 GB of local
instance storage, 64-bit platform, $0.50 per hour
High-CPU Medium Instance 1.7 GB,, 5 ECUs, 350 GB of local instance
storage, 32-bit platform, $0.17 per hour

(comment: one ECU provides the equivalent CPU capacity of a 1.0-1.2
GHz 2007 Opteron or 2007 Xeon processor.)

==============================================================================

Azure
------------------------------------------------------------
Extra Small 768 MB 1.0 GHz 20 GB Instance Storage $0.05
Small 1.75 GB 1.6 GHz 225 GB Instance Storage $0.12
Medium 3.5 GB 2 x 1.6 GHz 490 GB Instance Storage $0.24
Large 7 GB 4 x 1.6 GHz 1,000 GB Instance Storage High $0.48
Extra Large 14 GB 8 x 1.6 GHz 2,040 GB Instance Storage $0.96

==============================================================================

Rackspace
-----------------------------------------------------------
256MB RAM 10GB Disk $0.015/hr.
512MB RAM 20GB Disk $0.03/hr.
1,024MB RAM 40GB Disk $0.06/hr.
2,048MB RAM 80GB Disk $0.12/hr.
4,096MB RAM 160GB Disk $0.24/hr.
8,192MB RAM 320GB Disk $0.48/hr.
15,872MB RAM 620GB Disk $0.96/hr.

(CPU configures are not unknown)

[Tapir]

IMHO, the current price model means it is over for app engine.

[Barry Hunter]

I know its rather cliché but they really are apples to oranges.

The GAE instance gives you more than a single VPS instance.

Most notably is the Free APIs. In effect you have free access to many
things that probably cost you elsewhere. And they dont use the
resources of your instance.
eg on EC2 you pay extra for a Memcache instance (elasticache).

GAE Instances have in effect a free Elastic Load Balancer (continuing
the EC2 comparison) in front of your app. They have Cloudfront bundled
in too (the edge-cache). The image manipulation API, is similar to
running another gearman instance coordinating dedicated software for
resizing iamges (again offloading your instance)

And for the most part you still need a sysadmin to coordinate EC2.

Its the old paas vs iaas argument.

> --
> You received this message because you are subscribed to the Google Groups "Google App Engine" group.
> To post to this group, send email to google-a...@googlegroups.com.
> To unsubscribe from this group, send email to google-appengi...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/google-appengine?hl=en.
>
>

[Tapir]

For a medium traffic website, I can only one amazon EC2 instance:

"Large Instance 7.5 GB, 4 ECUs, 850 GB of local instance storage, 64-
bit platform, $0.34 per hour"

or

"High-Memory Extra Large Instance, 17.1 GB, 6.5 ECUs, 420 GB of local
instance storage, 64-bit platform, $0.50 per hour"

with self-installed memory cache and without using load balancer at
all.

For the same computing power, I need at least 4 gae backends, which
will cost more than 4 times money than ec2.
And the storage price is less than half of gae.

On Sep 2, 4:41 pm, Barry Hunter <barrybhun...@gmail.com> wrote:

[Joshua Smith]

It's about value. I've done EC2. It's not S3. S3 is easy like GAE. EC2 is much harder. And the learning curve to construct an EC2-based system that survives an outage is immense. (Witness Foursquare and Quora which both blew it.)

For my enterprise apps, the new GAE pricing is very reasonable. I'm saving something like $65K/year by not needing a sysadmin to set up and maintain these systems. The 3-year free ride was great, but I'm getting way more than a few hundred dollars of value each month. We'll eventually port to HR to get multi-threads.

For my free app, it's problematic. For Android game back ends, it's problematic. For many .com startups, it's problematic (which is not good for google, because having GAE as a startup incubator was a really good thing).

But for Enterprise apps, which is what GAE has been longing to get to, the new pricing really is just fine.

If you don't think so, then go use EC2. I don't think google is targeting you if you think you know how to build a distributed system, and you think you want to. They are targeting people who don't have the first clue how to build a distributed system, and those who have built them and know how hard it really is.

-Joshua

On Sep 2, 2011, at 5:20 PM, Tapir wrote:

> For a medium traffic website, I can only the amazon EC2

> "Large Instance 7.5 GB, 4 ECUs, 850 GB of local instance storage, 64-
> bit platform, $0.34 per hour"

> or

> "High-Memory Extra Large Instance, 17.1 GB, 6.5 ECUs, 420 GB of local
> instance storage, 64-bit platform, $0.50 per hour"

> with self-installed memory cache and without using load balancer at
> all.

> And the storage price is less than half of gae.
>
> For the same computing power, I need at least 4 gae backends, which
> will cost more than 4 times money than ec2.
>
>
> On Sep 2, 4:41 pm, Barry Hunter <barrybhun...@gmail.com> wrote:

[Barry Hunter]

Well yes. If your app could live on one EC2 instance, then yes you
will probably save there.

But if have serious traffic, EC2 is a lot of work.
try juggling say 200 EC2 instances. Webservers, Memcache nodes,
Database nodes, load balancers, dns management, backup nodes,
centralized file storage. Code deployment infestructure.
Monitoring/failver management. It adds up.
Even doing it via rightscale or cloud-formation will take you your
time to micromanage it.

Multi-datacenter (regions in EC2) failover? - good luck with that.

[Tapir]

But the target projects for app engine are big websites?

In fact, if I have a website as you described, I will hire an aws
admin and put the website on aws.
Yes, one aws admin is enough.

[Tapir]

When the development of my website is done, I will move to aws for
sure under the new gae price model.
I select gae to host my app in development process is for gae provide
some free quotas.

"They are targeting people who don't have the first clue how to build
a distributed system, and those who have built them and know how hard
it really is. "

If this is right, then I think the market space for gae will be very
small, under the new gae price model.

[GR]

Not sure I feel great about posting this.... because I haven't tried it myself, and it seems a bit too good to be true... but have you ever thought of using the new 4GH service from GoDaddy? It seems as though it's basically hassle free like GAE -- scaling up as needed.  Check out the "Features" and the "4GH vs. The Rest" tabs for more information.

Again.. not sure if posting this is embarassing or not.. but here's the link:

http://www.godaddy.com/hosting/web-hosting.aspx?ci=9009

[Barry Hunter]

Oh yes. they are going to let you push 10 Terabits/sec of data for
£4.99 a month.

Is this mysql database going to scale to 2000QPS? About terabytes of data?

... there are lots of great hosting packages around, ones that will
work perfectly well for small or even medium traffic.

> --
> You received this message because you are subscribed to the Google Groups
> "Google App Engine" group.

> To view this discussion on the web visit
> https://groups.google.com/d/msg/google-appengine/-/tsSQeRw1ek4J.

[GR]

haha.. yeah... I'm sure it's not going to scale up to those levels for sure... but for the small guys who are hosting mildly to moderately popular websites/services that don't fit within the free package on GAE, this may be a decent option.  

For my situation, it may work well -- 10 QPS with spikes up to probably 200 QPS.  The app I've got isn't overly huge and won't take long to port -- if it works, it will save me about $150/mo.  In my case it's worth a shot.

[Tapir]

Could you provide more information to show how you save $150/mo?
How many pageviews? How much will you cost when you use gea competitor
products?

[Anders]

That looks interesting. And no messy instances to have to pay huge amounts of money for. No Java support though. And I wonder how well their MySQL solution scales.

[Tapir]

Hi, this thread is just to compare computing prices.
GAE storage will cost extra money.

[Raymond C.]

Do you realize that you can do everything with the EC2 instances?  For example, you can use the extra 1.5GB memory of "Small Instance" on EC2 for a much more reliable memcache that you dont need to be afraid to be cleared by Google at any time.

Also, most API of GAE are NOT FREE.  You are not just paying for the instances.  Check you billing history if you really are using GAE.

[Andrew Cassidy]

I agree that one admin would be enough. I have not used EC2 myself but building a load-balancing, fault-tolerant cluster is relatively trivial if you've done it before.

I've just designed and built a prototype for hosting django using readily available free software.

[Anders]

Ok, sorry about that. But one thing to remember is that SQL is much more powerful than the GAE datastore when it comes to functionality which would demand a lower price for GAE. On the other hand the GAE datastore is designed for scaling, which may be difficult to achieve using ordinary relational databases like MySQL. I like the GAE datastore because it scales automatically. It's just having to pay for frontend instances I think really sucks.

[Tapir]

"Ok, sorry about that. But one thing to remember is that SQL is much

more ..."
hi, Anders,
again, this thread is to compare computing prices, not data storage.
You will pay extra data storage money when using gae.

I just want to get a reason from googlers why gae computing cost 10
times than competitors.

[Anders]

I have started a new thread about a possible reason for the higher prices: http://code.google.com/appengine/forum/?place=topic%2Fgoogle-appengine%2F6WqUi10P8Yo%2Fdiscussion

Just a speculation, but you can check it out in case it has some truth to it.

[Barry Hunter]

Mainly because GAE is not a 'computing' platform as such. If want raw
computing power then again elsewhere will probably suit you better
(and save some money)

A GAE instance is much more than a 'slice of cpu and memory' - it has
a whole 'platform' surrounding it.

Apples != Oranges again.

Or to put it another way, its like comparing two different cars on
their engine alone. You could get a average car, with a really good
engine (and be happy). Or you could get a more expensive car, with a
OK engine, but with nice suspension, interior and a sound system.

The two cars are good in different ways - need to look at the whole
car, to decide if does the job for you.

On Sat, Sep 3, 2011 at 10:59 PM, Tapir <tapi...@gmail.com> wrote:
> "Ok, sorry about that. But one thing to remember is that SQL is much
> more ..."
> hi, Anders,
> again, this thread is to compare computing prices, not data storage.
> You will pay extra data storage money when using gae.
>
> I just want to get a reason from googlers why gae computing cost 10
> times than competitors.
>
> Anders wrote:
>> Ok, sorry about that. But one thing to remember is that SQL is much more
>> powerful than the GAE datastore when it comes to functionality which would
>> demand a lower price for GAE. On the other hand the GAE datastore is
>> designed for scaling, which may be difficult to achieve using ordinary
>> relational databases like MySQL. I like the GAE datastore because it scales
>> automatically. It's just having to pay for frontend instances I think really
>> sucks.
>

[Tapir]

"A GAE instance is much more than a 'slice of cpu and memory' - it has
a whole 'platform' surrounding it. "

yes, but the surroundings will cost extra money!

[Francois Masurel]

Hi Tapir,

If you don't need the surroundings, GAE is probably not be the right solution for you.

Francois

[Tapir]

Yes, I need.
But this should NOT be reason for why gae computing charge 10 times
than competitors.
If it should, then the surroundings should not charge any more.

[Wilson MacGyver]

I wouldn't use the word "trivial". tell that to foursquare, quora, etc.

to setup something that's comparable to GAE, you'd have to setup
instances in both
amazon East and West, handle replication from both sides. have fail over.

and no, it's not just as easy as adding machines from both east and
west together as a cluster.
amazon charges you network fee for any traffic outside of a single
region. so ie, you pay for
any traffic between your instances in east and west. Plus you have to
deal with latency.

I understand people are upset about GAE pricing, but let's not pretend
this is as easy as
1: sign up for amazon, 2: setup 3 instances in amazon east, 3: webscale :)

On Sat, Sep 3, 2011 at 3:53 AM, Andrew Cassidy <an...@bytz.co.uk> wrote:
> I agree that one admin would be enough. I have not used EC2 myself but building a load-balancing, fault-tolerant cluster is relatively trivial if you've done it before.
>
> I've just designed and built a prototype for hosting django using readily available free software.
>

--
Omnem crede diem tibi diluxisse supremum.

[Tapir]

On Sep 6, 12:47 am, Wilson MacGyver <wmacgy...@gmail.com> wrote:
> I wouldn't use the word "trivial". tell that to foursquare, quora, etc.
>
> to setup something that's comparable to GAE, you'd have to setup
> instances in both
> amazon East and West, handle replication from both sides. have fail over.
>
> and no, it's not just as easy as adding machines from both east and
> west together as a cluster.
> amazon charges you network fee for any traffic outside of a single
> region. so ie, you pay for
> any traffic between your instances in east and west. Plus you have to
> deal with latency.

I don't what you are saying, app engine doesn't charge this? Please
show the document url.
AWS will not charge you for reading data in the same region, but app
engine will.

>
> I understand people are upset about GAE pricing, but let's not pretend
> this is as easy as
> 1: sign up for amazon, 2: setup 3 instances in amazon east, 3: webscale :)

Firstly, it is not hard, not harder than learning gae specified
things.
Secondly, we admit both services have their own advantages.
Thirdly, AWS can't do all app engine can do, but app engine can't do
all AWS can do.
Fourth, we are upset is not for app engien new price is high but for
it is TOO high!

[Jeff Schnitzer]

On Mon, Sep 5, 2011 at 9:47 AM, Wilson MacGyver <wmac...@gmail.com> wrote:
>
> I understand people are upset about GAE pricing, but let's not pretend
> this is as easy as
> 1: sign up for amazon, 2: setup 3 instances in amazon east, 3: webscale :)

Part of the problem is that there is a distinct difference between GAE
frontends and GAE backends. GAE frontends may or may not be
overpriced - certainly they provide compelling benefits WRT
scalability, fault tolerance, access to services, etc.

GAE backends, on the other hand, are clearly overpriced. They provide
little benefit over a server instance in AWS or rackspacecloud or
linode, and they cost 5X as much. The best argument you can make for
backends is that they have direct access to the GAE APIs, but you can
easily enough proxy those calls from outside.

Jeff

[Andrew Cassidy]

I wasn't trying to say 'You can replicate GAE on AWS trivially", GAE is a massive beast that does one thing in particular that I haven't experimented with yet: automatically creating and destroying running instances. I was trying to say you could host your own scaling application on any infrastructure you like with a little know how.

You can host in multiple data centers using Anycast addresses, you can round-robin your load balancers using clusterip or DNS, you can handle load balancer failures using basic ip failover mechanisms.

As long as you use some form shared storage backend for session data, such as memcache, you don't have to worry about sticky sessions on the load balanacers.

I have a working prototype on a basic testing network using (nothing amazon here)

• nginx as the load balancer and to serve static content
  
• uwsgi container for the applications
• mysql cluster for redundant data storage
• mysql proxy to load balance across the database servers without any code modification
• memcached for session storage and output caching
• corosync/pacemaker for ip failover
• rabbitmq/celery for task queueing

As I say, it's only a working prototype, and in most cases you could host a smallish application on a smallish vps anywhere. Most VPS providers now provide fault-tolerance and failover as standard. I'm moving my application to one such host.

Using AWS for fault-tolerance, I just read this blog post: http://blog.rightscale.com/2008/03/26/setting-up-a-fault-tolerant-site-using-amazons-availability-zones/

If anyone wants to fund an experiment on setting up an application on AWS I'd be more than happy to try it.

[Pol]

On Sep 7, 2:53 am, Andrew Cassidy <an...@bytz.co.uk> wrote:
> I wasn't trying to say 'You can replicate GAE on AWS trivially", GAE is a
> massive beast that does one thing in particular that I haven't experimented
> with yet: automatically creating and destroying running instances. I was
> trying to say you could host your own scaling application on any
> infrastructure you like with a little know how.

Anyone can write a web page, an iPhone app or even drive a car on a
race track with "a little know how". That's doesn't you won't crash at
the first "serious" problem.

Look, we can debate all day long on the perceived value of GAE is and
what a fair price for the service, but putting this type of service
together does require quite a bit of know how.

- Pol

[Andrew Cassidy]

--
You received this message because you are subscribed to the Google Groups "Google App Engine" group.

To post to this group, send email to google-a...@googlegroups.com.
To unsubscribe from this group, send email to google-appengi...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/google-appengine?hl=en.

Your post indicates you did not read the first sentence of mine.

--
Andrew Cassidy BSc (Hons) MBCS

http://www.jonisdumb.com/

[will]

This is very good information. Any pointer in utilizing Anycast? Any
VPS support it?

On Sep 7, 2:53 am, Andrew Cassidy <an...@bytz.co.uk> wrote:

> I wasn't trying to say 'You can replicate GAE on AWS trivially", GAE is a
> massive beast that does one thing in particular that I haven't experimented
> with yet: automatically creating and destroying running instances. I was
> trying to say you could host your own scaling application on any
> infrastructure you like with a little know how.
>
> You can host in multiple data centers using Anycast addresses, you can
> round-robin your load balancers using clusterip or DNS, you can handle load
> balancer failures using basic ip failover mechanisms.
>
> As long as you use some form shared storage backend for session data, such
> as memcache, you don't have to worry about sticky sessions on the load
> balanacers.
>
> I have a working prototype on a basic testing network using (nothing amazon
> here)
>

>    - nginx as the load balancer and to serve static content
>    - uwsgi container for the applications
>    - mysql cluster for redundant data storage
>    - mysql proxy to load balance across the database servers without any
>    code modification
>    - memcached for session storage and output caching
>    - corosync/pacemaker for ip failover
>    - rabbitmq/celery for task queueing

>
> As I say, it's only a working prototype, and in most cases you could host a
> smallish application on a smallish vps anywhere. Most VPS providers now
> provide fault-tolerance and failover as standard. I'm moving my application
> to one such host.
>

> Using AWS for fault-tolerance, I just read this blog post:http://blog.rightscale.com/2008/03/26/setting-up-a-fault-tolerant-sit...

[Andrew Cassidy]

Due to the nature of anycasting, it's not something I've been able to experiment with myself.

Here's a good post on anycasting: http://blog.rootshell.be/2008/12/23/dnsbe-more-anycasting/

In terms of VPS providers, gandi.net say they've done it before http://wiki.gandi.net/en/hosting/faq-fr-us#i-have-my-own-asn-and-assigned-24-for-anycast-purposes-can-i-use-a-gandi-vps-to-host-my-anycast-service-assuming-that-i-use-other-providers-to-expand-the-geographic-and-network-scope